US hyperscale data centers emit about as much CO2 as Denmark, study estimates

Fossil fuels are still the biggest source of electricity for hyperscale data centers in the US, according to a new paper in the journal Nature Sustainability, as reported by The Verge. The study focuses on the massive facilities, spanning millions of square feet, that developers are racing to build in large part to support generative AI.
The team, led by Gianluca Guidi, a postdoctoral fellow in the Department of Biostatistics at Harvard T.H. Chan School of Public Health, used satellite imagery to identify 403 hyperscale data centers in the US. They estimated each facility's power consumption from its electrical capacity. Then, using Environmental Protection Agency data, they approximated how much carbon pollution each center would have produced if connected to the local power grid between May 2024 and April 2025.
The totals are estimates, not measurements. Altogether, the 403 facilities likely consumed around 82 terawatt-hours of electricity that year, about 1.8 percent of all US electricity consumption. The electricity demand produced 25.7 million metric tons of CO2, comparable to the annual carbon footprint of Denmark, according to the paper. The reason is the power mix: about 54 percent of the electricity the facilities consumed likely came from gas and coal.
The country comparisons beyond Denmark come from John Sterman, director of the MIT System Dynamics group and MIT Sloan Sustainability Initiative, who was not involved in the research. He says the 82 TWh is more than about 148 nations, a majority of the world's countries, including Austria, Israel and Venezuela but excluding the largest economies such as the US and China, each use per year. He also says the emissions exceed the total carbon emissions of about 113 nations. Sterman disclosed that he used Claude to help compare the paper's findings with countries' electricity use and emissions. "I think AI can be extraordinarily helpful," he says, but the harms must be limited, especially because they are irreversible: once the CO2 is emitted, it cannot be undone. He says policymakers should require data center operators to bring new clean energy online.
The study also shows how much the local electricity mix matters. Generating electricity in heavily coal-reliant Wyoming led to roughly seven times more carbon emissions per kilowatt-hour than in hydropower-rich Oregon. Guidi writes that two data centers using exactly the same amount of electricity can have dramatically different carbon footprints simply because of where they are located.
Fengqi You, a professor in energy systems engineering at Cornell University who was also not involved in the study, says the results highlight the problem of "long-term carbon lock-in". Power plants and supporting infrastructure built to meet fast-growing data center demand may operate for decades, so decisions made today could shape emissions well into the future.
Both You and Guidi pointed out that researchers must go to great lengths to estimate energy use because data center operators won't disclose it. Guidi and his colleagues are using their research to build a website that tracks the environmental impact of individual data centers. A bipartisan bill making its way through Congress could also mandate greater national disclosure of data centers' electricity use. Gina McCarthy, former White House national climate advisor and EPA administrator, said in a statement shared by the authors that communities and policymakers need reliable, independent data on what these facilities mean for air, health and the climate, and called the study an important step.
The article lists limits. The study does not account for local air pollution from new gas infrastructure built to meet data center demand, or for gas turbines that some AI companies have installed to power themselves without connecting to the grid. Hyperscale data centers also cover more than AI-related computing, and they are just a fraction of the thousands of data centers in the US, with more appearing each year.
Key facts
- A Nature Sustainability paper led by Gianluca Guidi of Harvard T.H. Chan School of Public Health used satellite imagery to identify 403 US hyperscale data centers.
- The facilities likely consumed around 82 TWh from May 2024 to April 2025, about 1.8 percent of all US electricity consumption.
- Their electricity demand produced an estimated 25.7 million metric tons of CO2, comparable to Denmark's annual footprint, with about 54 percent of the power likely from gas and coal.
- Coal-reliant Wyoming produced roughly seven times more CO2 per kilowatt-hour than hydropower-rich Oregon, so location changes the footprint of identical data centers.
- The study excludes local air pollution from new gas infrastructure and off-grid gas turbines, and operators' refusal to disclose consumption forces researchers to estimate.
Why it matters
Hyperscale data centers are being built at speed, in large part to support generative AI, and this paper puts a number on their carbon cost. The estimated 82 TWh is about 1.8 percent of all US electricity consumption, and the 25.7 million metric tons of CO2 is comparable to Denmark's annual footprint. The main driver is the grid: about 54 percent of the electricity likely came from gas and coal. Guidi argues that AI's footprint is not a fixed property of the technology; the same computing infrastructure has very different consequences depending on the electricity system behind it. Fengqi You adds the lock-in concern: plants built now may run for decades.
Who it affects
Data center developers and operators, whose siting and power choices drive the result. Policymakers and communities near new facilities, who according to Gina McCarthy need reliable, independent data on air, health and climate effects. Regions with coal-heavy grids, such as Wyoming in the study's comparison, versus hydropower-rich ones such as Oregon. Researchers, who have to estimate consumption because operators won't disclose it.
How to use it
This is a study, not a tool, but it points to practical levers. Guidi says there are choices about where to build and how to power the infrastructure. Sterman says policymakers should require data center operators to bring new clean energy online. Guidi and colleagues are building a website to track the environmental impact of individual data centers, and a bipartisan bill in Congress could mandate greater national disclosure of data centers' electricity use.
How solid is it
The paper is peer-reviewed in Nature Sustainability, and the lead author is at Harvard T.H. Chan School of Public Health. The method is indirect: satellite imagery to find 403 facilities, electrical capacity to estimate power use, and EPA data to approximate emissions as if each center were connected to the local grid. The 82 TWh and 25.7 million metric ton figures are estimates. The comparisons with 148 and 113 nations come from John Sterman, who was not involved in the study and who disclosed that he used Claude to make the comparison. Fengqi You, also not involved, is quoted on the implications.
Risks and caveats
The estimates do not include local air pollution from new gas infrastructure, or gas turbines that some AI companies installed to power themselves off-grid. Hyperscale data centers cover more than AI-related computing, so the figures should not be read as AI's footprint alone. They are also just a fraction of the thousands of data centers in the US. Because operators don't disclose consumption, the numbers rest on capacity-based estimates. The article gives no timescale for the proposed clean-energy requirement or for the bill's passage.
“I actually find that encouraging, because it means these impacts aren't inevitable. We have choices about where we build and how we power this infrastructure.”
— Gianluca Guidi, lead author, Harvard T.H. Chan School of Public Health